Acedly AI on Microsoft Teams: Real-Time Interview Copilot for Teams (2026)
How Acedly AI works on Microsoft Teams — invisible to screen sharing, compatible with Teams live captions and recording, tuned for enterprise interview loops. Verified stealth checks for your next Teams round.
Devon Park
Head of Research, Acedly
What is a Microsoft Teams interview assistant?
A Microsoft Teams interview assistant is a desktop copilot built specifically for the Teams capture model. It is not a generic AI chat with a Teams skin. The category exists because Teams routes audio differently from Zoom, exposes different surfaces during screen share, and ships its own AI features (live captions, Copilot summaries, intelligent recap) that change what a candidate has to think about.
The job is the same as any AI interview assistant — listen, ground a response in your résumé and the JD, and render the draft on a screen the interviewer cannot see — but the implementation details that make it work on Teams are not the same as the ones that make it work on Zoom. A tool that demos cleanly on Zoom can still appear in a Teams "share window" picker, get pulled into a Together-mode background, or trip a Teams Premium watermark. The audit is platform-specific.
Why Microsoft Teams dominates enterprise interview loops
If you graph interview platforms by company size, Teams and Zoom essentially trade places at the Fortune 1000 line. Below it, Zoom dominates — startups, mid-market tech, most boutique firms. Above it, the picture flips. Anyone on a Microsoft 365 E3 or E5 tenant — which is the default for large enterprises — runs interviews on Teams by policy, because IT has already paid for it and procurement has already vetted it.
Concretely, Teams is the default for:
- Investment banking and finance superdays. Goldman, JPMorgan, Morgan Stanley, BlackRock — all heavy Microsoft 365 shops.
- Big Four consulting and the MBB strategy firms when interviewing with corporate clients (the firms themselves often use Zoom internally, but client-facing case rounds default to whatever the client uses, and the client almost always uses Teams).
- Fortune 500 corporate functions — finance, ops, supply chain, legal, internal product roles at non-tech companies.
- Microsoft and the Microsoft partner ecosystem for obvious reasons.
- UK and EU enterprises more uniformly than the US, where Teams penetration is higher per capita.
Zoom is still the default for product, design, and engineering interviews at companies founded after roughly 2015. The honest split is that if you're interviewing for a senior corporate role at a mature enterprise, you're almost certainly on Teams, and the stealth audit your assistant did against Zoom does not transfer.
How an AI interview assistant works on Microsoft Teams specifically
The pipeline — audio capture, streaming transcription, grounded inference, hidden render — is conceptually the same on every platform. The implementation differences worth knowing are concentrated in three places.
Teams' audio routing is not Zoom's
Teams uses Microsoft's Real-time Media Stack, which on Windows routes meeting audio through audiodg.exe (the Windows Audio Device Graph isolation process) before it reaches your output device. A serious copilot subscribes to the system loopback at the WASAPI layer rather than trying to hook the Teams process, because hooking process-specific audio is brittle across Teams desktop updates and breaks entirely on Teams web. On macOS, Teams' audio path through the system mixer is more conventional; the equivalent capture is via Core Audio's tap on the default output device.
Live captions, Together mode, and gallery views
Teams ships three features that Zoom doesn't have at the same fidelity, and each one is worth understanding:
- Live captions are generated by Microsoft's own server-side speech-to-text from the meeting audio stream. They never see your local UI. If your interviewer turns on live captions, they're transcribing what they hear, not what's on your screen.
- Together mode composites participants onto a shared virtual background. The compositing happens server-side from each participant's camera feed. It does not pull from your active windows. (Some early Teams beta features tried to ingest desktop content for "scene" detection; these did not ship to general availability.)
- Large gallery / Large meeting views behave the same way — they render camera feeds, not screen content, unless the candidate explicitly shares.
Teams desktop vs. Teams web
The Teams desktop app on Windows and Mac is a native shell around an Electron-rendered UI. The capture surfaces — the windows visible in the screen-share picker, the focus model, the Alt-Tab behaviour — are OS-level concerns and behave like any other native app. Teams on the web (teams.microsoft.com) runs entirely in the browser. Screen sharing on Teams web uses the browser's getDisplayMedia API, which means the share picker is the browser's picker, not Teams'. The set of windows the interviewer can see is determined by Edge or Chrome, and a copilot that's invisible on Teams desktop should also test against the browser-level capture model on Teams web.
The Teams-specific stealth checklist
A copilot is either invisible on Teams or it isn't, and "invisible on Zoom" is not a sufficient proxy. Run through these six items explicitly:
- Excluded from Teams' "share window" picker. When the candidate clicks Share → Window, the Acedly window does not appear in the list of shareable windows. This requires
WDA_EXCLUDEFROMCAPTUREon Windows andNSWindowSharingNoneon macOS, set on the actual render surface, not just the parent. - Invisible when sharing the entire desktop. If the candidate shares the whole screen, the assistant's pixels are not in the captured frame. This is the same OS flag, but tested against full-screen capture rather than per-window.
- Hidden from Teams' Together mode and Large gallery composition. These features pull from camera feeds, not desktop content, so the risk is largely theoretical — but the test is to start a Together-mode call with a colleague, alt-tab through your desktop, and confirm nothing leaks.
- Excluded from Teams meeting recording. Teams recordings are composed server-side from the streams the meeting is receiving (camera + screen-share feed). Local UI that wasn't being shared is not in the recording. The verification is to record a test meeting, share a window, alt-tab to Acedly, and review the recording in Stream / OneDrive.
- Compatible with Teams Premium, Teams for Government, and Teams Rooms. Teams Premium adds watermarking and intelligent recap; neither one ingests local screen content beyond what's being shared. Teams for Government runs in a separate cloud (GCC / GCC High / DoD) but uses the same desktop client. Teams Rooms is a different model entirely — these are conference-room appliances, and a candidate is unlikely to be on the appliance side, but mixed-mode meetings (where one participant is in a Teams Room) still respect per-participant capture isolation.
- Survives the "share computer audio" toggle. When the candidate enables "share computer audio" during screen share, Teams routes audio differently and, in older Teams desktop versions, briefly re-evaluated capture surfaces. Modern Teams clients (post-2024) handle this cleanly, but the test is worth running explicitly.
Comparison: Acedly vs. browser-tab copilots vs. generic AI on Teams
| Feature | Acedly | Teams web extensions | Browser-tab AI | Generic AI chat |
|---|---|---|---|---|
| Median end-to-end latency | ~98 ms | ~600–900 ms | ~500–900 ms | ~2–4 seconds |
| Stealth on Teams desktop | Yes (OS-level capture exclusion) | N/A — no desktop surface | No — visible in window list | No — visible in window list |
| Stealth on Teams web (Edge / Chrome) | Yes — verified against getDisplayMedia | Visible to extension host | Tab-only, fragile | Not designed for it |
| Reads coding-interview UI on screen | Yes — Coderpad, HackerRank, etc. | Limited | Limited | No |
| Grounded in résumé + JD | Yes by default | Sometimes | Sometimes | Only if pasted in |
| Compatible with Teams Premium tenants | Yes | Depends on extension policy | Depends on browser policy | Yes |
The "Teams web extensions" column matters because some assistant products ship as browser extensions designed to run inside Teams web. These are visible to the Teams web JavaScript host, which means they show up in extension permission audits and, on managed devices, in the IT admin console. They are not stealth in the same sense an OS-level desktop copilot is.
What to verify 10 minutes before your Teams interview
The dry run takes ten minutes and prevents the failure modes that actually happen:
- Do a test share with a colleague using both "share window" and "share desktop." This is the single most important test. Acedly should be absent from the window picker and absent from the captured frame.
- Confirm Teams hasn't auto-elevated to a Teams Rooms call. Mixed-mode meetings with a Teams Room device on the other end have stricter capture and watermark behaviour. The candidate's local capture isolation is unchanged, but the test is worth running.
- Verify your hotkey works against Teams' focus model. Teams steals keyboard focus for in-call shortcuts (Ctrl+Shift+M for mute, etc.). Make sure your assistant's hotkey doesn't collide and that it fires regardless of which window has focus.
- Move Acedly to a second monitor if you have one. This is belt-and-suspenders rather than a stealth requirement — the OS-level exclusion holds regardless — but a second monitor reduces the chance of a careless reveal during the call.
- Disable Windows notifications and Focus Assist. Teams interviews routinely get interrupted by Outlook calendar pop-ups. Focus Assist or "Do Not Disturb" eliminates this independently of Acedly.
Live captions, Copilot, and what Microsoft's AI sees
The honest version: Teams now ships Microsoft 365 Copilot meeting summaries, live captions, and "intelligent recap" in many enterprise tenants. Candidates ask, reasonably, whether any of these features can detect or transcribe what's happening on their local screen.
The answer is no, with one caveat. Microsoft 365 Copilot in Teams ingests the meeting transcript (server-side STT of the meeting audio) and any documents shared via the meeting chat or screen share. It does not have access to your local desktop, your other windows, or anything outside what's actually being transmitted to the meeting. Live captions work the same way — they transcribe the meeting's audio stream, not your microphone or your screen. Intelligent recap operates on the post-call recording.
The caveat is that whatever ends up in the recording or transcript is now indexed by Copilot for that tenant. If a candidate's audio includes them reading an Acedly draft verbatim — which is the cadence failure we warn about in the main pillar — that text is now in a searchable transcript that the company's recruiter can retrieve later via Copilot. The risk isn't detection; it's the candidate's own mouth.
Teams Premium adds watermarking on shared content and prevents some recording vectors, but again, the watermark is on the shared frame, not on the candidate's local desktop. Teams for Government runs in isolated clouds with stricter compliance posture but the same Teams client and the same capture isolation rules.